Mass Deployment of Neural Networks: Proof-of-Concept with Screening of Intracranial Hemorrhage Using an Open Dataset

Background and Purpose: Intracranial hemorrhage (ICH) is considered an emergency that requires rapid medical or surgical management. Previous studies have utilized artificial intelligence to attempt to automate and expedite the diagnosis of this pathology on neuroimaging. However, these studies have utilized local, institution-specific data for training of these networks that limits the deployment of neural networks across broader hospital networks or regions due to data biases. We demonstrate the creation and validation of a neural network based on an openly available imaging data tested on data from our institution demonstrating a high-efficacy, institution-agnostic network. Methods: A retrospective study was completed after institutional review board approval. A dataset was created from de-identified, publicly available non-contrast computed tomography (CT) images of known ICH. This data was used to train a neural network using distinct windowing parameters and custom random augmentation. This network was then validated in two phases using cohort-based (phase 1) and longitudinal (phase 2) approaches. Results :Our CNN was trained on 752,807 openly available slices, which included 112,762 slices containing intracranial hemorrhage. In phase 1, final network performance for intracranial hemorrhage showed a receiver operating characteristic curve (AUC) was 0.99. At the inflection point, our model showed a sensitivity of 98% at a threshold specificity of 99%. In phase 2, we obtained an AUC of 0.99 after analysis of 726 scans with a negative predictive value of 99.70% (n=726). Conclusions: To our knowledge, this is the first report that demonstrates an effective neural network trained on completely open data for screening ICH at an unrelated institution. This study demonstrates a proof-of-concept for widespread deployment of screening neural networks to multiple sites, while maintaining high efficacy. Further studies are necessary in multiple hospital system types, with different patient populations, to understand site-specific characteristics that may limit network applicability. Funding Statement: None. Declaration of Interests: No relevant disclosures to report and no conflicts of interest. Ethics Approval Statement: A retrospective study was completed after institutional review board approval.